This AI Startup Recommends Disease Therapies Based on Your Genome

The healthcare industry is advancing rapidly, with new drugs developed and innovative medical technologies coming to the market each year. Yet, there are still over 12 million patients misdiagnosed every year in the US. Oriel Research, an AI startup previously in stealth mode until today, is tackling this dire problem with their novel machine learning algorithms coupled with the principles of precision medicine.

Oriel Research’s mission is to diagnose diseases and recommend matching therapies using machine learning models. It was founded in late 2017 by Eila Arich-Landkof, and Israeli entrepreneur following her pursuits of computer science, bioinformatics and genomic research.

[gallery link="none" columns="1" size="full" ids="2724"]

Eila Arich-Landkof began her career in software engineering at multinational corporation R&D centers in Israel, such as Elbit Systems, Cisco Systems and Microsoft. She later took on research roles in the labs of Jaenisch and Erlich at the MIT Whitehead Institute conducting tissue cultures, extracting DNA and working at the bench. She was later recruited as an Associate Researcher at the Getz Lab, located at the Broad Institute of MIT and Harvard University, where she worked on making clinical data easily accessible for patients. In fact, a poster describing her work achieved a 1st place ranking in the Broad’s scientific retreat competition. Although unconventionally, her background in AI in self-taught entirely thanks to the myriad of online resources available today, however her studies in computer science and computational mathematics lend itself well to the realm. The Oriel Research team is composed of kindred minds from the Broad and Weizmann Institutes; Dr. Noam Shoresh, Dr. Keren Yizhak, and Dr. Ben-Hamo. Additionally, clinician Dr. Lior Braunstein from Sloan Kettering, and Software Architect Nir Bar-El.

[gallery link="none" columns="1" size="full" ids="2725"]

The basic unit of biology is a cell, and diseases are caused by cells that are not working properly. For example, healthy cells have mechanism that ensure they don’t divide without control (so as not cause damage to the entire body), but in the case of cancer, these mechanisms are broken. It is not always easy to tell whether a cell is healthy or not and there are thousands of factors that determine how a cell works. Although key among these factors is the DNA, the set of instruction for building a human that is present in every human cell (except red blood cells). These instructions take the form of a very long molecular “word” made up of only four letters (the nucleic acids, denoted by the letters ‘A’,’C’,’T’, and ‘G’). Over the last few decades, the industry has experienced amazing progress in sequencing technologies, technologies that allow us to read the precise sequence of letters of the DNA in a cell. All the cells in the body of an individual carry almost the same set of instructions. ‘Almost’ because there are processes (such as cell division, or exposure to radiation) that lead to slight changes, or mutations, in the DNA of different cells. Thus, for example, when the DNA in a cell acquires a set of mutations that cause it to start proliferating without control, that cell may become a malignant, cancerous tumor.

Sick people are treated based on symptoms that give the physician clues about what is wrong with them. The problem? Often two patients presenting the same symptoms may respond very differently to the same treatment. The reason is that despite the similarities, there are differences at the molecular level (sometimes at the level of the DNA) between the malfunctioning cells in the two individuals. The promise of precision/personalized medicine is that by reading the DNA (and more) in the diseased cells of each  patient, it will be possible to tailor a treatment that is specifically matched to the state of that patient's cells.

For example, consider a patient suffering from a malignant and aggressive melanoma cancer where standard therapies and immunotherapy have no effect on the cancer. But after adopting precision medicine principles and checking the patient for a BRAF mutation, the patient was a carrier and thus a candidate for the BRAF inhibitor therapy, a novel targeted therapy. However, not all BRAF mutation carrier patients will benefit from such a therapy. Predicting who might benefit from such a treatment and which patient might not is accomplished by Oriel Research’s Machine Learning models.

[gallery link="none" columns="1" size="full" ids="2726"]

Oriel Research is tackling precision medicine through a novel innovative approach. They use machine learning classifiers to diagnose and recommend how to treat diseases, all based on molecular profiles, and their approach is attempting to further transcend precision medicine. Oriel Research identifies disease therapies according to genomic and molecular profiling. For instance, consider three patients suffering from lung cancer, skin cancer and leopard syndrome. All three patients are prescribed three different treatment based on blood tests, CT and PET scans. However, all three patients may exhibit similar molecular and genomic profiles, which is effectively another perspective by which to classify patients. Through clustering patients according to their genomic and molecular profiles, Oriel Research effectively derives new sub-disease classifications. And with these new sub-disease classifications, they recommend disease therapies that match the molecular and genomic profile properties and exhibit the same response to therapy.

Oriel Research collects a massive amount of data (genetic and other) about patients and healthy individuals, as well as clinical information about treatments that patients received and their outcome. They apply deep learning classification neural networks to find the features that are connected by the success of a particular treatment. This allows them to group the patients by traditional standards of diagnosis of a disease, as well as by the likely successful treatments based on their individual molecular profile.

Additionally, the startup provides a data harmonization service to its academia and industry clients. There’s no standard format for genomic data files and analysis, and preparing the data for analysis is an effort of many orders of magnitude for effective algorithmic work. Among their client base is The Polak Lab at the Mount Sinai Hospital in New York.

Oriel Research is a super young startup but already generating revenues, and its cloud activity was granted for sponsorship from Google. In the future, the vision for Oriel Research is to provide disease identification from a single blood drop and therapy matching recommendations for doctors who don’t have the bandwidth to deal with rapid technological advances. “We want to democratize data-driven healthcare for doctors and patients” explained Eila.

To this day, Eila maintains the argument that the internet lacks a central library to learn about diseases and therapy effectiveness. And so Oriel Research is her journey to bring data to the patients in need.

Outside of Oriel Research, Eila can be found leading a monthly meetup called Deep Learning in Production in New York City or Cambridge with over 2,000 members, or periodically speaking at Google AI events where she aims to bridge the gap between data and technology in the industry. You can watch the meetup’s content on youtube here, or subscribe to the group. Eila can be reached at [email protected] or follow her on twitter at @eilalan1.